Transcript CMAQ Modeling

A Brief Introduction to CMAQ
Serena H. Chung
BioEarth Working Group 1 Seminar
May 21, 2012
Outline
•
•
•
•
•
Chemical Transport Models (CTMs)
CMAQ Model Components
CMAQ Output
Parallel Programming in CMAQ
WRF and CMAQ Linkages
Chemical Transport Models (CTMs)
• Transport:
– Same physics as numerical weather model, but different numerical methods
are needed
• Chemistry
– Focuses on criteria pollutants which negatively affect human health
• Ozone (O3): plant stresser  ecosystem impact
• Particular Matter (PM) in air quality community or aerosols in climate science
community
– Consists of hundreds if not thousand of chemical species
– Climate impact: scatter and absorb radiation; affects cloud formation
• NOx (=NO + NO2): most of which eventually deposits as nitrate  ecosystem
impact
• SO2 : forms, sulfate aerosol, contributes to acidification ecosystem impact
• Mercury and other air toxics
Chemical Transport Model Equation
•
Solves for species concentration Cs using mass conservation equation for each grid
cell and time step:
¶Cs ¶Cs ¶Cs
¶Cs ¶ æ ¶Cs ö ¶ æ ¶Cs ö ¶ æ ¶Cs ö
+u
+v
+w
= çK x
÷ + çKy
÷ + Rs + Ds + Es
÷ + çKz
¶t
¶x
¶y
¶z ¶x è ¶x ø ¶y è ¶y ø ¶z è ¶z ø
horizontal
change in advection
concentration
•
vertical
advection
horizontal
diffusion
chemical emission
vertical
reaction
diffusion
deposition
Input or derived from numerical weather model (e.g., WRF, MM5)



Wind fields: u, v, w
Eddy diffusivity (turbulent diffusion) coefficients: Kx=Ky, Kz
Temperature, Pressure, (& Radiation Fields):



Clouds & Precipitation:



To calculate reaction rates
Emissions rate can also be temperature and/or light dependent
Aqueous-phase reactions
Removal rate by wet deposition
Dry deposition velocities vd,s, where Ds = vd,s Cs,layer 1
Chemical Mechanisms
• A chemical mechanism is a condensed set of chemical reactions
– Chosen to represent conditions of interest, .e.g, O3 in polluted environment,
stratospheric O3
•
Example - University of Leeds Master Chemical Mechanism
– Thousands of species and >10,000 chemical reactions
•
Options in CMAQ v5.0
– CB05:
– SAPRC99:
– SAPRC07:
~72 species, ~187 reactions
~88 species, ~144 reactions
~150 species, ~413 reactions
HNO3
PAN
Nitrogen cycle in the
troposphere is tightly
coupled to O3 & aerosol
chemistry
OH●
hn
DMS or VOC
NO
NO2
O3
NO3●
N2O5
O3
NO2 + Aer
R can be lots of stuff with
carbon and hydrogen
atoms
HNO3
H2O
●
●
RO2 or HO2
NOx (NO+NO2)
Atmospheric
Deposition
Aerosol Size Distribution
Volume
Distribution
Number
Distribution
Typical Urban Conditions
Based on Whitby, Atmos. Environ., 1978
Aerosol Size Distribution & Composition
Volume
Distribution
Number
Distribution
Typical Urban Conditions
Based on Whitby, Atmos. Environ., 1978
Aerosol Size Distribution
Volume
Distribution
Number
Distribution
Typical Urban Conditions
Based on Whitby, Atmos. Environ., 1978
Aerosol Size Distribution
Volume
Distribution
Number
Distribution
Typical Urban Conditions
Based on Whitby, Atmos. Environ., 1978
Aerosol Size Distribution:
Number vs Surface vs Volume
• Number
Number
– Affects the number of cloud
droplets that form
• Surface Area
– Affects the amount of
radiation that is scatter or
absorbed
Surface Area
• Volume
– Portional to mass, used by the
National Ambient Air Quality
Standards (NAAQS)
– PM10 & PM2.5 standards
designed to distinguish coarse
and fine particles.
Volume
Figure 7.6
Seinfeld & Pandis
2.5 mm
10 mm
Aerosol Size Representations
•
•
No size representation, simulate only aerosol mass
Use few lognormal distributions (e.g, CMAQ uses 3), each characterized by
– Total particle number concentrations
– Median diameter
– Geometric standard deviation
•
Use sectional bins
– Track aerosol mass only, or
– Track aerosol number and mass
•
Mixtures
– Internally mixed – all particles within a bin or lognormal distribution have the same
chemical composition
– Externally mixed – each particle contains one “species”, so species are not mixed
– Combination of the two
•
Effective number of species Neff for sectional bins with number and mass: Neff
= (1 + Nspecies) Nmixture Nbin
Nspecies = ~ 20
Nmixture = 1-5
Nbin = 4-30
Chemical Tranport Model
¶Cs ¶Cs ¶Cs
¶Cs ¶ æ ¶Cs ö ¶ æ ¶Cs ö ¶ æ ¶Cs ö
+u
+v
+w
= çK x
÷ + çKy
÷ + Rs + Ds + Es
÷ + çKz
¶t
¶x
¶y
¶z ¶x è ¶x ø ¶y è ¶y ø ¶z è ¶z ø
horizontal
change in advection
concentration
•
vertical
advection
horizontal
diffusion
chemical emission
vertical
reaction
diffusion
deposition
Operator splitting -- the equation is split into parts and solved separately:
1)
2)
3)
4)
5)
6)
7)
vertical diffusion, emission, & dry deposition
horizontal advection
vertical advection
horizontal diffusion
cloud processes (includes aqueous chemistry)
gas-phase chemistry
aerosol chemistry
Horizontal Discretization in CMAQ
AIRPACT-3 Example:
12-km x 12-km grids in
Lambert Conformal Conic Projection
Arakawa C Grid
vi,j+1
j+1
Ci,j,s
ui+1,j
j
North
Dy
East
j-1
i-1
i
Dx
i+1
Vertical Discretization in CMAQ
WRF Example: Terrain-Following, Hydrostatic Pressure Grid
h=
ph - pht
phs - pht
where Ph = hydrostatic pressure
Pressure at model top: pht ~ 10,000 Pa (~ 15 km)
~30-40 levels with first layer height at ~ 40 m
wi,k+1
Ci,k,s
k+1
ui+1,j
k
Up
Dh
East
k-1
i-1
i
Dx
i+1
Figure not to scale
Adapted from Figure 2.1 of Skamarock et al., 2008
Vertical Discretization
AIRPACT-4 Example
CMAQ Grid Cell in 3-Dimension
wi,j,k+1
vi,j+1,k
ui+1,j,k
ui,j,k
Up
vi,j/2,k
North
wi,j,k
East
•
•
•
•
Air density
Temperature
Pressure
Water mixing ratios
(vapor, rain, snow, ice)
• Gas- and aerosol-phase
chemical species mixing
ratios
Why does CMAQ take so long to run?
•
The nature of chemical transport models:
– Gas phase: ~ 100 chemical species
– Particle phase: ~20 species, 3-16 size bins
 effectively ~60-320 species minimum
•
ODEs governing the chemical reactions:
– Nonlinear
– Stiff -- eigenvalues of Jacobian : negative; min/max ratio is ~ 109
Figure from Gustafason et al. (2005)
(http://www.mmm.ucar.edu/wrf/users/workshop/WS2005
/presentations/sessions8/4-Gustafson.pdf
Model Time Steps
• WRF:
– Physics: recommendation is 6 seconds per km of
Dx, i.e., 72 seconds for 12-km x 12-km grids
– Radiation: recommendation is 1 minute per km of
Dx, i.e., 12 minutes for 12-km x 12-km grids
• CMAQ:
– Synchronization between all processes: ~ 1-3 min
– Adaptive time step within each process
CMAQ Model Components
http://www.airqualitymodeling.org/cmaqwiki/index.php?title=File:Figure5-1.png
CMAQ Model Components
• Meteorological fields from a
numerical weather model
• Usually MM5 or WRF,
though other models
can also be used
Meteorology
Example of Layer 1
Temperature and Wind
Fields from WRF
http://www.airqualitymodeling.org/cmaqwiki/index.php?title=File:Figure5-1.png
http://www.atmos.washington.edu/mm5rt
CMAQ Model Components
• Converts WRF or MM5
output files into CMAQready files
• Calculates/diagnoses
parameters not provided by
WRF (e.g., Monin-Obukhov
length)
• Calculates dry deposition
velocities (depends on landuse type and turbulence
characteristics)
• Keeps the same horizontal
grid cell size
• Collapses WRF layers into
fewer layers if desired
http://www.airqualitymodeling.org/cmaqwiki/index.php?title=File:Figure5-1.png
CMAQ Model Components
Emissions: Various
models/processors, e.g.,
Transportation
Industrial
Residential
Power Plants
Fire
Biogenic
etc
http://www.airqualitymodeling.org/cmaqwiki/index.php?title=File:Figure5-1.png
CMAQ Model Components
Initial Conditions:
• Usually from a previous run
• Only ~ 2-3 days for spin-up
required
http://www.airqualitymodeling.org/cmaqwiki/index.php?title=File:Figure5-1.png
CMAQ Model Components
Boundary Conditions Using:
• “Idealized’ profile,
• Results from a coarser,
bigger domain CMAQ
simulation, or
• Results of global CTMs
http://www.airqualitymodeling.org/cmaqwiki/index.php?title=File:Figure5-1.png
CMAQ Model Components
Photolysis Rate Calculations
• Using look-up table for
clear-sky conditions and
adjusted “online” based on
cloud conditions
http://www.airqualitymodeling.org/cmaqwiki/index.php?title=File:Figure5-1.png
CMAQ Model Components
Solves
¶Cs ¶Cs ¶Cs
¶C ¶ æ ¶C ö ¶ æ ¶C ö ¶ æ ¶C ö
+u
+v
+ w s = ç K x s ÷ + ç K y s ÷ + ç K z s ÷ + Rs + Ds + Es
¶t
¶x
¶y
¶z ¶x è ¶x ø ¶y è ¶y ø ¶z è ¶z ø
http://www.airqualitymodeling.org/cmaqwiki/index.php?title=File:Figure5-1.png
CMAQ Output
• Hourly, 3-dimensional concentrations (.e.g, parts
per billion or mg m-3) of chemical species
• Hourly accumulated wet and dry deposition (.e.g,
kg ha-1 hr-1) for relevant species
• netCDF files
– same as WRF, but different conventions for date/time
– read/write easier with use of Models-3 I/O API library
• Examples:
– http://lar.wsu.edu/airpact
– http://lar.wsu.edu/airpact/gmap/testC.html
CMAQ Output : AIRPACT Example
12-km, Surface-Layer,
Hourly Concentrations of
Secondary Organic Aerosl (SOA)
• Lots of stuff at:
– AIRPACT-3: http://lar.wsu.edu/airpact
– AIRPACT-4: http://lar.wsu.edu/airpact/gmap/testC.html
CMAQ Output: Vertical Distribution
AIRPACT-4 Output for
10AM PST on Feb 23, 2011
O3 Concentation
Parallel Progamming in CMAQ
• Distributed Memory using Message Passing Interface (MPI) (WRF supports
•
OpenMP and MPI)
Divide and conquer by horizontal domain decomposition
– Similar to WRF, but specifics are different
12 13 14 15
•
8
9
10 11
4
5
6
7
0
1
2
3
For I/O, each processor gets the data for its subdomain by extracting the data
from the full domain. However, only one processor is responsible for writing to
the output data files; thus, gathering full domain data is required before
writing
WRF-CMAQ Soft Link
Static
Geographical Data
GEOGRID
Global
Data
METGRID
Emission
Models
UNGRIB
Geographical & Large-scale
Meteorological Data
Interpolated to simulation grids
MCIP
REAL
ICON
Initial & Boundary
Conditions
WRF
CCTM
BCON
Meteorological
Fields
JPROC
Coupled WRF-CMAQ
Static
Geographical Data
GEOGRID
Global
Data
METGRID
UNGRIB
Geographical & Large-scale
Meteorological Data
Interpolated to simulation grids
Emission
Models
REAL
MCIP
Initial & Boundary Conditions
WRF
call aqprep
call cmaq_driver
call feedback_read
Meteorological
Fields
ICON
CCTM
Speciated
Aerosol Size
Distributions, &
O3
Concentrations
BCON
JPROC
WRF-CMAQ Domains
CMAQ_COL_DIM
delta_x
CMAQ_ROW_DIM
5 columns
5 rows
CMAQ Domain
Max CMAQ Domain
delta_y
WRF Domain
Adapted from Figure 2 of Wong et al., Geosci. Model Dev., 2012
Coupled WRF-CMAQ Computaional
Performance
Table 1 of Wong et al., Geosci. Model Dev., 2012
WRF only
MCIP
Offline CMAQ
Loose couple system, Total time
Coupling system w/o feedback and call frequency ratio 5:1
Coupling system w/ feedback and call frequency ratio 5:1
Execution time
CAM
RRTMG
0:19:59
0:18:50
0:02:31
0:02:31
1:18:28
1:19:05
1:40:58
1:40:26
1:41:12
1:48:59
1:43:39
2:54:25
Table 2 of Wong et al., Geosci. Model Dev., 2012
Processor
configuration
w/o
feedback
CAM
speedup
w/
feedback
4x8
8x8
8x16
2:05:06
1:19:46
0:55:28
2:08:21
1:21:57
0:55:12
1.57
2.26
speedup
1.57
2.33
w/o
feedback
RRTMG
Speedup
w/
feedback
2:13:17
1:24:12
0:56:38
3:19:25
1:58:21
1:14:14
1.58
2.35
Based on 24-hour simulations for a 12-km eastern US domain
speedup
1.68
2.69
Some resources
•
•
•
•
•
http://cmaq-model.org
http://cmascenter.org/
Seinfeld, J.H. and S.N. Pandis, Atmospheric Chemistry and Physics: From Air
Pollution to Climate Change, John Wiley & Sons, 2006.
Jacob, D.J., Introduction to Atmospheric Chemistry, Princeton University Press,
1999.
Jacobson, M.Z., Fundamentals of Atmospheric Modeling, Cambridge University
Press, 1999